My NHL model is breaking down

**InTheRed** · 02-04-11, 10:53 AM

I feel that the preseason projections should be less important as the season progresses. As we are well beyond halfway through the season, those projections to me seem irrelevant. A system based on what some "experts" think a player will do, is sketchy to start off but the further into a season you go, the less impact those numbers should have on your system. Those numbers are based off of information gathered in July and Aug. We're now in Feb. 6 months is a long time, a lot of factors can be found as to why those numbers aren't relevant.

For the Devils, no one, I mean no one, thought they would be in last place in the East on Feb 2nd. But they are. Your projection numbers are probably based off of a Devils team projected to be in the top 4 of the league. To me, that would completely throw them out of my system until the team and the players get back to a "projected" range.

I would try weighting your model a bit as you progress through the season. projections are just guesses, and now you're 6 months into those guesses. Of course they are going to be off. To me, unless they are statistically relevant (on pace), I wouldn't even consider using them this far into the season.

**Justin7** · 02-04-11, 11:03 AM

Are devils players differing greatly from their projections?

**element1286** · 02-04-11, 11:08 AM

Originally posted by Justin7

Are devils players differing greatly from their projections?

For the most part they are playing much worse than their projections.

**element1286** · 02-04-11, 11:25 AM

Using only my preseason projections I have NJ -123. Using only my in season stats I have NJ +125. Using my full model I have NJ +108.

**Justin7** · 02-04-11, 01:12 PM

I am not an NHL handicapper, so be careful with any ideas I offer.

How much do NHL stats tend to revert to the mean? If these NJ players are playing terribly, are they likely to continue playing this bady? What measuring stick are you using for the players?

Are NHL player performances somewhat constant, or do they predictably fluctuate? Is a season average a bettor predictor than the last 10 games?

How is the market moving regarding your NJ plays? If NJ opened at +130 and closed at +150, your model gave you the right side of the opener at least when it projected +200.

**Inkwell77** · 02-04-11, 03:43 PM

You might know this, but they have a new coach and are playing a completely different system in these past 10 games, maybe that is it? They were playing an uptempo run and gun type of game and now they are playing a more puck control based game.

**element1286** · 02-04-11, 05:30 PM

Originally posted by Justin7

I am not an NHL handicapper, so be careful with any ideas I offer.

How much do NHL stats tend to revert to the mean? If these NJ players are playing terribly, are they likely to continue playing this bady? What measuring stick are you using for the players?

Are NHL player performances somewhat constant, or do they predictably fluctuate? Is a season average a bettor predictor than the last 10 games?

How is the market moving regarding your NJ plays? If NJ opened at +130 and closed at +150, your model gave you the right side of the opener at least when it projected +200.

I'm using GVT (goals versus threshold) for the players. It is similar in theory to WAR, in that it attempts to define a player versus a standard replacement level. Which allows me to translate a lineup into a theoretical per game goals for and goals against for each.

NHL players tend to revert to the mean. Like baseball it is more accurate for players in their prime with a fair amount of experience. And it is sketchy for young players, and players at the end of their careers. I don't have any exacts on that though. In my experience season average is a better predictor than the past 10 games.

Market isn't moving much in either direction, so it is hard to gauge.

I think my main problem is that it puts doubt into my other numbers. If I see I have a 15-20 cent advantage on a line, then I question whether my line is correct or not. I'm probably just going to lay off the Devils for a while till the numbers get back in order.

**element1286** · 02-04-11, 05:31 PM

Originally posted by Inkwell77

You might know this, but they have a new coach and are playing a completely different system in these past 10 games, maybe that is it? They were playing an uptempo run and gun type of game and now they are playing a more puck control based game.

I am aware, but that isn't quantifiable, or at least I don't have the numbers to make any predictions. But thanks for the input.

**element1286** · 02-04-11, 05:32 PM

Line is moving in my direction now, but that has to be because Hedberg is the confirmed starter tonight, and not Broduer.

**gohabsgo** · 02-05-11, 12:06 AM

They took Lemaire out of his retirement and now that the team his playing accordingly to his system expect NJ to win more. You should add a Lemaire constant to your calculations.

**Peregrine Stoop** · 02-05-11, 08:45 AM

Originally posted by element1286

I am aware, but that isn't quantifiable, or at least I don't have the numbers to make any predictions. But thanks for the input.

you should probably just stay away from that team for 'til your projections are somewhat close

**Miz** · 02-05-11, 11:09 AM

Even though your projections are off for that team (recently), how has the line moved from open to close? toward or against your line? Has there been a fair amount of movement on these games' lines from open to close?

How has your model done historically in other seasons versus the line movement?

Do you use a decaying function to phase out your preseason projections? For young players that don't have much data do you just weight that toward league average (with that being the decayed initial value, as opposed to a player specific historical mean)?

I should also add that I don't have much experience modeling hockey. Nor do I keep up with it during the year.